Abstract
Background: Implementation of telemental health care in emergency departments (EDs) in the United States (U.S.) has been increasing. Artificial intelligence (AI) can augment traditional qualitative research methods; little is known about its efficiency and accuracy. This study sought to understand ED directors' qualitative recommendations for improving telemental health care implementation and to understand how AI could facilitate analysis of qualitative survey responses. Methods: Directors at a nationally representative sample of 279 U.S. EDs that used telemental health care completed an open-ended survey question about improving telemental health care implementation between June 2022 and October 2023. Two groups of researchers completed independent qualitative coding of responses: one group used traditional qualitative methods, and one group used AI (ChatGPT 4.0) to facilitate analysis. Both groups independently developed a codebook, came to consensus on a combined codebook, and each group independently used it to code the survey responses. The two groups identified themes in ED directors' recommendations and compared codebooks and code application across traditional and AI approaches. Results: Themes included (1) recommendations for improving telemental health care directly and (2) recommendations for improving mental health care systems broadly to make telehealth more effective. ED directors' most common recommendation was enabling faster and more streamlined access to telemental health care. AI augmented human coding by identifying two valid codes not initially identified by human analysts. In codebook application, 75% of responses were coded consistently across AI and human coders. Conclusions and Relevance: For US EDs using telemental health care, there is a need to improve timeliness and efficiency of access to telemental health care.